A tool that transforms how populations can be tracked
For generations, fishermen insisted the sea was not silent — and science, with its customary caution, was slow to believe them. Researchers at the University of Victoria have now built the first systematic catalog of fish vocalizations, using underwater hydrophone arrays and machine learning to identify eight species by sound alone with 88% accuracy. The work, conducted in the waters of Barkley Sound, British Columbia, offers more than a technical achievement: it suggests that the ocean has been speaking all along, and that conservation may be transformed by the simple act of learning to listen.
- Traditional fish surveys rely on nets, traps, and divers — methods that are costly, disruptive, and fundamentally at odds with the goal of protecting the animals being studied.
- Two species, canary rockfish and vermillion rockfish, had never been documented making sounds before this study, revealing how much of underwater life remains uncharted even in well-studied ecosystems.
- A machine learning model trained on 47 acoustic characteristics — pitch, duration, frequency — correctly matched sounds to species 88% of the time, even between closely related fish.
- A striking physical pattern emerged: smaller fish produce higher-pitched calls, larger ones lower tones, hinting that sound alone could reveal not just who is present, but how big they are.
- Researchers are now working toward counting fish populations by call frequency and investigating whether regional fish populations carry distinct acoustic dialects — a 'Boston accent' beneath the waves.
Fishermen have long claimed to hear fish communicating underwater, a notion the scientific community received with polite skepticism. A new study from the University of Victoria suggests those fishermen were right.
The research team deployed a hydrophone array in Barkley Sound, British Columbia, synchronizing acoustic recordings with video footage to match sounds to the fish producing them. Over the course of the study, they catalogued more than 1,000 individual vocalizations from eight species living on rocky reefs — including rockfish, lingcod, pile perch, and kelp greenling. Two of those species, canary rockfish and vermillion rockfish, had never been recorded making sounds before.
Feeding the recordings into a machine learning model trained on 47 acoustic properties, the team achieved 88% accuracy in identifying species by sound alone. They also began to decode meaning: certain calls corresponded to courtship, aggression, feeding, and escape. A consistent physical pattern emerged — smaller fish produced higher-pitched sounds, larger fish lower ones — suggesting that acoustic data could reveal not just species identity, but individual size.
The implications for conservation are significant. Passive hydrophone arrays can monitor continuously without nets, traps, or divers, leaving fish and their behaviors undisturbed. Looking ahead, the team hopes to estimate population density through call frequency and to explore whether fish from different regions vocalize differently — regional dialects in the deep. The project represents something larger than a new monitoring tool: an invitation to fundamentally expand how humanity listens to the living world.
Fishermen have long claimed to hear fish talking underwater. For years, the scientific establishment mostly dismissed these accounts with polite skepticism, the kind that comes with a mental eye-roll once the storyteller has left the room. A new study suggests those fishermen may have been onto something real.
Researchers at the University of Victoria set out to test the premise seriously. They deployed an underwater acoustic localization array in Barkley Sound, British Columbia—essentially a network of hydrophones capable of triangulating sound sources beneath the surface with precision. By synchronizing this acoustic data with video footage, they could match specific underwater noises to the fish producing them. The work, published in the Journal of Fish Biology, represents the first systematic effort to build a reliable catalog of fish vocalizations tied to identified species.
Over the course of their study, the team recorded more than 1,000 individual fish sounds. They focused on eight species found on rocky reefs: several varieties of rockfish, lingcod, pile perch, and kelp greenling. Two of these—canary rockfish and vermillion rockfish—had never been documented making sounds before, officially joining the roster of vocal marine life. The researchers then fed their recordings into a machine learning model trained to recognize 47 distinct acoustic characteristics: pitch, duration, frequency patterns, and other measurable properties of each call.
The model proved remarkably accurate. It correctly identified which species had made a given sound 88 percent of the time, even when distinguishing between closely related fish that might seem acoustically similar to human ears. The team also began to decode what these sounds meant. Some calls correlated with known behaviors—courtship displays, aggressive encounters, feeding activity, escape responses. A pattern emerged in the physics of it all: smaller fish produced higher-pitched sounds, larger fish lower ones, a relationship that suggested acoustic monitoring could potentially reveal not just what species were present, but how big they were.
This opens a door that marine conservation has long needed. Traditional fish surveys require nets, traps, or divers—methods that are invasive, expensive, and labor-intensive. They also disturb the animals being studied. Acoustic monitoring, by contrast, is passive. A hydrophone array can listen continuously, day and night, through seasons and years, without touching a single fish or altering its behavior. If scientists can reliably estimate species, size, location, and abundance from sound alone, they gain a tool that transforms how populations can be tracked.
The team's next steps are already taking shape. They want to develop methods for counting fish based on call frequency—essentially, using the density and pattern of vocalizations to estimate how many individuals occupy a given area. They're also curious about something more whimsical: whether fish from different regions have distinct acoustic dialects, regional variations in how they vocalize. The possibility of discovering a Boston accent in the ocean, as one researcher put it, captures something true about what this work represents—a fundamental expansion of how we listen to the natural world, and what we might learn if we actually pay attention.
Notable Quotes
If scientists can reliably estimate fish size, their locations, and how many of them there are in a spot just by listening, they'll gain a new, non-invasive conservation tool.— Research team findings
The Hearth Conversation Another angle on the story
Why did it take so long for scientists to take fish sounds seriously?
Part of it was just the barrier to observation. You can't hear fish underwater without specialized equipment. Fishermen heard them because they were out there, listening, but anecdotes aren't data. The technology had to catch up to the claim.
What makes the machine learning part important here?
The model doesn't just recognize sounds—it learns the acoustic fingerprint of each species. It can distinguish between eight different fish even when their calls might sound similar to us. That's the difference between a neat observation and a usable tool.
You mentioned size matters. How does that change things?
If you can hear a fish and know its species, the pitch of its call tells you roughly how big it is. Suddenly you're not just counting fish, you're getting their size distribution. That's crucial for understanding whether a population is healthy or aging out.
The non-invasive part seems like the real breakthrough.
It is. Nets and divers stress the fish, they're expensive to deploy repeatedly, and they only give you a snapshot. A hydrophone array can listen for months or years without disturbing anything. You get a continuous picture of what's actually happening.
What about these regional accents they want to study?
That's still speculative, but if it's real, it would mean fish populations have cultural variation—learned vocal patterns passed down within groups. That would be profound for how we think about fish as social beings.